mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
503 lines
49 KiB
HTML
503 lines
49 KiB
HTML
|
||
|
||
<!DOCTYPE html>
|
||
<html class="writer-html5" lang="en" data-content_root="../">
|
||
<head>
|
||
<meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />
|
||
|
||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||
<title>TensorRT-LLM Model Weights Loader — tensorrt_llm documentation</title>
|
||
<link rel="stylesheet" type="text/css" href="../_static/pygments.css?v=80d5e7a1" />
|
||
<link rel="stylesheet" type="text/css" href="../_static/css/theme.css?v=e59714d7" />
|
||
<link rel="stylesheet" type="text/css" href="../_static/copybutton.css?v=76b2166b" />
|
||
|
||
|
||
<script src="../_static/jquery.js?v=5d32c60e"></script>
|
||
<script src="../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
|
||
<script src="../_static/documentation_options.js?v=5929fcd5"></script>
|
||
<script src="../_static/doctools.js?v=9a2dae69"></script>
|
||
<script src="../_static/sphinx_highlight.js?v=dc90522c"></script>
|
||
<script src="../_static/clipboard.min.js?v=a7894cd8"></script>
|
||
<script src="../_static/copybutton.js?v=65e89d2a"></script>
|
||
<script src="../_static/js/theme.js"></script>
|
||
<link rel="index" title="Index" href="../genindex.html" />
|
||
<link rel="search" title="Search" href="../search.html" />
|
||
</head>
|
||
|
||
<body class="wy-body-for-nav">
|
||
<div class="wy-grid-for-nav">
|
||
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
|
||
<div class="wy-side-scroll">
|
||
<div class="wy-side-nav-search" >
|
||
|
||
|
||
|
||
<a href="../index.html" class="icon icon-home">
|
||
tensorrt_llm
|
||
</a>
|
||
<div role="search">
|
||
<form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
|
||
<input type="text" name="q" placeholder="Search docs" aria-label="Search docs" />
|
||
<input type="hidden" name="check_keywords" value="yes" />
|
||
<input type="hidden" name="area" value="default" />
|
||
</form>
|
||
</div>
|
||
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="Navigation menu">
|
||
<p class="caption" role="heading"><span class="caption-text">Getting Started</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../overview.html">Overview</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../quick-start-guide.html">Quick Start Guide</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../key-features.html">Key Features</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../release-notes.html">Release Notes</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">Installation</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../installation/linux.html">Installing on Linux</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../installation/build-from-source-linux.html">Building from Source Code on Linux</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../installation/windows.html">Installing on Windows</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../installation/build-from-source-windows.html">Building from Source Code on Windows</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../installation/grace-hopper.html">Installing on Grace Hopper</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">LLM API</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../llm-api/index.html">API Introduction</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../llm-api/reference.html">API Reference</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">LLM API Examples</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../llm-api-examples/index.html">LLM Examples Introduction</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../llm-api-examples/customization.html">Common Customizations</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../llm-api-examples/llm_api_examples.html">Examples</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">Model Definition API</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.layers.html">Layers</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.functional.html">Functionals</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.models.html">Models</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.plugin.html">Plugin</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.quantization.html">Quantization</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.runtime.html">Runtime</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">C++ API</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../_cpp_gen/executor.html">Executor</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../_cpp_gen/runtime.html">Runtime</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">Command-Line Reference</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../commands/trtllm-build.html">trtllm-build</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../commands/trtllm-serve.html">trtllm-serve</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">Architecture</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="overview.html">TensorRT-LLM Architecture</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="core-concepts.html">Model Definition</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="core-concepts.html#compilation">Compilation</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="core-concepts.html#runtime">Runtime</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="core-concepts.html#multi-gpu-and-multi-node-support">Multi-GPU and Multi-Node Support</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="checkpoint.html">TensorRT-LLM Checkpoint</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="workflow.html">TensorRT-LLM Build Workflow</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="add-model.html">Adding a Model</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">Advanced</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/gpt-attention.html">Multi-Head, Multi-Query, and Group-Query Attention</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/gpt-runtime.html">C++ GPT Runtime</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/executor.html">Executor API</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/graph-rewriting.html">Graph Rewriting Module</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/inference-request.html">Inference Request</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/inference-request.html#responses">Responses</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/lora.html">Run gpt-2b + LoRA using GptManager / cpp runtime</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/expert-parallelism.html">Expert Parallelism in TensorRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/kv-cache-reuse.html">KV cache reuse</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/speculative-decoding.html">Speculative Sampling</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../advanced/disaggregated-service.html">Disaggregated-Service (experimental)</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">Performance</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../performance/perf-overview.html">Overview</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../performance/perf-benchmarking.html">Benchmarking</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../performance/perf-best-practices.html">Best Practices</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../performance/perf-analysis.html">Performance Analysis</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">Reference</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../reference/troubleshooting.html">Troubleshooting</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../reference/support-matrix.html">Support Matrix</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../reference/precision.html">Numerical Precision</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../reference/memory.html">Memory Usage of TensorRT-LLM</a></li>
|
||
</ul>
|
||
<p class="caption" role="heading"><span class="caption-text">Blogs</span></p>
|
||
<ul>
|
||
<li class="toctree-l1"><a class="reference internal" href="../blogs/H100vsA100.html">H100 has 4.6x A100 Performance in TensorRT-LLM, achieving 10,000 tok/s at 100ms to first token</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../blogs/H200launch.html">H200 achieves nearly 12,000 tokens/sec on Llama2-13B with TensorRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../blogs/Falcon180B-H200.html">Falcon-180B on a single H200 GPU with INT4 AWQ, and 6.7x faster Llama-70B over A100</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../blogs/quantization-in-TRT-LLM.html">Speed up inference with SOTA quantization techniques in TRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../blogs/XQA-kernel.html">New XQA-kernel provides 2.4x more Llama-70B throughput within the same latency budget</a></li>
|
||
</ul>
|
||
|
||
</div>
|
||
</div>
|
||
</nav>
|
||
|
||
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="Mobile navigation menu" >
|
||
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
|
||
<a href="../index.html">tensorrt_llm</a>
|
||
</nav>
|
||
|
||
<div class="wy-nav-content">
|
||
<div class="rst-content">
|
||
<div role="navigation" aria-label="Page navigation">
|
||
<ul class="wy-breadcrumbs">
|
||
<li><a href="../index.html" class="icon icon-home" aria-label="Home"></a></li>
|
||
<li class="breadcrumb-item active">TensorRT-LLM Model Weights Loader</li>
|
||
<li class="wy-breadcrumbs-aside">
|
||
<a href="../_sources/architecture/model-weights-loader.md.txt" rel="nofollow"> View page source</a>
|
||
</li>
|
||
</ul>
|
||
<hr/>
|
||
</div>
|
||
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
|
||
<div itemprop="articleBody">
|
||
|
||
<section id="tensorrt-llm-model-weights-loader">
|
||
<h1>TensorRT-LLM Model Weights Loader<a class="headerlink" href="#tensorrt-llm-model-weights-loader" title="Link to this heading"></a></h1>
|
||
<section id="overview">
|
||
<h2>Overview<a class="headerlink" href="#overview" title="Link to this heading"></a></h2>
|
||
<p>The weights loader is designed for easily converting and loading external weight checkpoints into TensorRT-LLM models.</p>
|
||
</section>
|
||
<section id="workflow">
|
||
<h2>Workflow<a class="headerlink" href="#workflow" title="Link to this heading"></a></h2>
|
||
<p>Weight checkpoints can be generated from all sources, and may have different naming and data layouts compared to TRT-LLM’s requirements. E.g.:</p>
|
||
<div class="highlight-bash notranslate"><div class="highlight"><pre><span></span><span class="c1"># HuggingFace LLaMA checkpoints</span>
|
||
<span class="o">{</span>
|
||
<span class="w"> </span><span class="s2">"model.embed_tokens.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>vocab_size,<span class="w"> </span>hidden_size<span class="o">])</span>
|
||
<span class="w"> </span><span class="s2">"model.layers.0.input_layernorm.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"model.layers.0.mlp.down_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size,<span class="w"> </span>inter_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"model.layers.0.mlp.gate_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>inter_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"model.layers.0.mlp.up_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>inter_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"model.layers.0.post_attention_layernorm.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"model.layers.0.self_attn.q_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"model.layers.0.self_attn.k_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"model.layers.0.self_attn.v_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"model.layers.0.self_attn.o_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span>...,
|
||
<span class="o">}</span>
|
||
<span class="c1"># TensorRT-LLM expected weights</span>
|
||
<span class="o">{</span>
|
||
<span class="w"> </span><span class="s2">"transformer.vocab_embedding.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>vocab_size,<span class="w"> </span>hidden_size<span class="o">])</span>
|
||
<span class="w"> </span><span class="s2">"transformer.layers.0.input_layernorm.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"transformer.layers.0.mlp.down_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size,<span class="w"> </span>inter_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"transformer.layers.0.mlp.gate_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>inter_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"transformer.layers.0.mlp.up_proj.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>inter_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"transformer.layers.0.post_layernorm.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span><span class="s2">"transformer.layers.0.attention.qkv.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size<span class="w"> </span>*<span class="w"> </span><span class="m">3</span>,<span class="w"> </span>hidden_size<span class="o">])</span>,<span class="w"> </span><span class="c1"># Different layout</span>
|
||
<span class="w"> </span><span class="s2">"transformer.layers.0.attention.dense.weight"</span>:<span class="w"> </span>torch.Tensor<span class="o">([</span>hidden_size,<span class="w"> </span>hidden_size<span class="o">])</span>,
|
||
<span class="w"> </span>...,
|
||
<span class="o">}</span>
|
||
</pre></div>
|
||
</div>
|
||
<p>Conversion means converting the dictionary of <code class="docutils literal notranslate"><span class="pre">{external_keys:external_weights}</span></code> into <code class="docutils literal notranslate"><span class="pre">{tllm_keys:tllm_weights}</span></code>, it includes changing the naming logic and data layouts, and is contains of the following parts:</p>
|
||
<ol class="arabic simple">
|
||
<li><p>Translate a TRT-LLM parameter name into external-format name(s).</p></li>
|
||
<li><p>Loading tensor slice(s) according to the translated names.</p></li>
|
||
<li><p>Postprocess the tensor(s) into target layout.</p></li>
|
||
</ol>
|
||
<section id="translator">
|
||
<h3>Translator<a class="headerlink" href="#translator" title="Link to this heading"></a></h3>
|
||
<p>TRT-LLM parameter names are translated in units of sections divided by dots. E.g.:</p>
|
||
<table class="docutils align-default">
|
||
<thead>
|
||
<tr class="row-odd"><th class="head text-center"><p>TensorRT-LLM key</p></th>
|
||
<th class="head text-center"><p><code class="docutils literal notranslate"><span class="pre">transformer</span></code></p></th>
|
||
<th class="head"><p>.</p></th>
|
||
<th class="head text-center"><p><code class="docutils literal notranslate"><span class="pre">layers</span></code></p></th>
|
||
<th class="head"><p>.</p></th>
|
||
<th class="head text-center"><p><code class="docutils literal notranslate"><span class="pre">0</span></code></p></th>
|
||
<th class="head"><p>.</p></th>
|
||
<th class="head text-center"><p><code class="docutils literal notranslate"><span class="pre">attention</span></code></p></th>
|
||
<th class="head"><p>.</p></th>
|
||
<th class="head text-center"><p><code class="docutils literal notranslate"><span class="pre">dense</span></code></p></th>
|
||
<th class="head"><p>.</p></th>
|
||
<th class="head text-center"><p><code class="docutils literal notranslate"><span class="pre">weight</span></code></p></th>
|
||
</tr>
|
||
</thead>
|
||
<tbody>
|
||
<tr class="row-even"><td class="text-center"><p>Translated external key</p></td>
|
||
<td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">model</span></code></p></td>
|
||
<td><p>.</p></td>
|
||
<td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">layers</span></code></p></td>
|
||
<td><p>.</p></td>
|
||
<td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">0</span></code></p></td>
|
||
<td><p>.</p></td>
|
||
<td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">self_attn</span></code></p></td>
|
||
<td><p>.</p></td>
|
||
<td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">o_proj</span></code></p></td>
|
||
<td><p>.</p></td>
|
||
<td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">weight</span></code></p></td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
<p>The mapping between TRT-LLM keywords and HF keywords are described in <code class="docutils literal notranslate"><span class="pre">tllm_to_externel_key_dict</span></code> of <code class="docutils literal notranslate"><span class="pre">ModelWeightsLoader</span></code> class object. <br />
|
||
If any of the mappings has one-to-multiple corresponding, the translated key will get multiplied accordingly. E.g.:</p>
|
||
<table class="docutils align-default">
|
||
<thead>
|
||
<tr class="row-odd"><th class="head text-center"><p>TensorRT-LLM key and related keyword mapping</p></th>
|
||
<th class="head text-center"><p>Translated external keys</p></th>
|
||
</tr>
|
||
</thead>
|
||
<tbody>
|
||
<tr class="row-even"><td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">transformer.layers.0.attention.qkv.weight</span></code><br><code class="docutils literal notranslate"><span class="pre">{"qkv":[q_proj,</span> <span class="pre">k_proj,</span> <span class="pre">v_proj]}</span></code></p></td>
|
||
<td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">model.layers.0.self_attn.q_proj.weights</span></code><br><code class="docutils literal notranslate"><span class="pre">model.layers.0.self_attn.k_proj.weights</span></code><br><code class="docutils literal notranslate"><span class="pre">model.layers.0.self_attn.v_proj.weights</span></code></p></td>
|
||
</tr>
|
||
<tr class="row-odd"><td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">transformer.layers.0.mlp.fc.weight</span></code><br><code class="docutils literal notranslate"><span class="pre">{"weight":[qweight,</span> <span class="pre">qzeros,</span> <span class="pre">scales]}</span></code></p></td>
|
||
<td class="text-center"><p><code class="docutils literal notranslate"><span class="pre">model.layers.0.mlp.gate_proj.qweight</span></code><br><code class="docutils literal notranslate"><span class="pre">model.layers.0.mlp.gate_proj.qzeros</span></code><br><code class="docutils literal notranslate"><span class="pre">model.layers.0.mlp.gate_proj.scales</span></code></p></td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
<p>The default <code class="docutils literal notranslate"><span class="pre">tllm_to_externel_key_dict</span></code> is based on HF LLaMA as:</p>
|
||
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">ModelWeightsLoader</span><span class="p">:</span>
|
||
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">model_dir</span><span class="p">,</span> <span class="n">customized_key_dict</span><span class="p">:</span> <span class="nb">dict</span> <span class="o">=</span> <span class="p">{})</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="o">...</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">tllm_to_externel_key_dict</span> <span class="o">=</span> <span class="p">{</span>
|
||
<span class="s2">"transformer"</span><span class="p">:</span> <span class="s2">"model"</span><span class="p">,</span>
|
||
<span class="s2">"vocab_embedding"</span><span class="p">:</span> <span class="s2">"embed_tokens"</span><span class="p">,</span>
|
||
<span class="s2">"lm_head"</span><span class="p">:</span> <span class="s2">"lm_head"</span><span class="p">,</span>
|
||
<span class="s2">"ln_f"</span><span class="p">:</span> <span class="s2">"norm"</span><span class="p">,</span>
|
||
<span class="s2">"attention"</span><span class="p">:</span> <span class="s2">"self_attn"</span><span class="p">,</span>
|
||
<span class="s2">"qkv"</span><span class="p">:</span> <span class="p">[</span><span class="s2">"q_proj"</span><span class="p">,</span> <span class="s2">"k_proj"</span><span class="p">,</span> <span class="s2">"v_proj"</span><span class="p">],</span>
|
||
<span class="s2">"dense"</span><span class="p">:</span> <span class="s2">"o_proj"</span><span class="p">,</span>
|
||
<span class="s2">"gate"</span><span class="p">:</span> <span class="s2">"up_proj"</span><span class="p">,</span>
|
||
<span class="s2">"proj"</span><span class="p">:</span> <span class="s2">"down_proj"</span><span class="p">,</span>
|
||
<span class="s2">"fc"</span><span class="p">:</span> <span class="s2">"gate_proj"</span><span class="p">,</span>
|
||
<span class="s2">"input_layernorm"</span><span class="p">:</span> <span class="s2">"input_layernorm"</span><span class="p">,</span>
|
||
<span class="s2">"post_layernorm"</span><span class="p">:</span> <span class="s2">"post_attention_layernorm"</span><span class="p">,</span>
|
||
<span class="p">}</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">tllm_to_externel_key_dict</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">customized_key_dict</span><span class="p">)</span>
|
||
<span class="o">...</span>
|
||
</pre></div>
|
||
</div>
|
||
<p>It can be updated through passing <code class="docutils literal notranslate"><span class="pre">customized_key_dict</span></code> when initializing <code class="docutils literal notranslate"><span class="pre">ModelWeightsLoader</span></code>.</p>
|
||
<p>The dictionary will also get updated according to the layer classes. When iterating over parameters,
|
||
if the layer class has attribute <code class="docutils literal notranslate"><span class="pre">tllm_to_externel_key_dict</span></code>, for keywords exist both in the default one and the layer-specified one,
|
||
the weight loader will translate according to the layer attribute with higher priority.
|
||
This can enable the support for different quantization precisions automatically.</p>
|
||
</section>
|
||
<section id="loading-function">
|
||
<h3>Loading function<a class="headerlink" href="#loading-function" title="Link to this heading"></a></h3>
|
||
<p>The loading function can load an arbitrary tensor slice according to its <code class="docutils literal notranslate"><span class="pre">key</span></code>, <code class="docutils literal notranslate"><span class="pre">tp_size</span></code>, <code class="docutils literal notranslate"><span class="pre">tp_dim</span></code> and <code class="docutils literal notranslate"><span class="pre">tp_rank</span></code>.</p>
|
||
<p>The template for loading function is as following.</p>
|
||
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">load_tensor</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">key</span><span class="p">,</span> <span class="n">tp_size</span><span class="p">,</span> <span class="n">tp_dim</span><span class="p">,</span> <span class="n">tp_rank</span><span class="p">):</span>
|
||
<span class="c1"># Retrieve file pointer index</span>
|
||
<span class="k">if</span> <span class="n">key</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">shard_map</span><span class="p">:</span>
|
||
<span class="n">ptr_idx</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">shard_map</span><span class="p">[</span><span class="n">key</span><span class="p">]</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="kc">None</span>
|
||
|
||
<span class="c1"># Load tensor from the corresponding shard</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">format</span> <span class="o">==</span> <span class="n">ModelWeightsFormat</span><span class="o">.</span><span class="n">SAFETENSORS</span><span class="p">:</span>
|
||
<span class="n">tensor</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">shards</span><span class="p">[</span><span class="n">ptr_idx</span><span class="p">]</span><span class="o">.</span><span class="n">get_slice</span><span class="p">(</span><span class="n">key</span><span class="p">)</span>
|
||
<span class="n">tensor_shape</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">get_shape</span><span class="p">()</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="o">...</span>
|
||
|
||
<span class="c1"># Shard and return a tensor slice</span>
|
||
<span class="n">slice_shape</span> <span class="o">=</span> <span class="o">...</span>
|
||
<span class="k">return</span> <span class="n">tensor</span><span class="p">[</span><span class="n">slice_shape</span><span class="p">]</span>
|
||
</pre></div>
|
||
</div>
|
||
<p>When initializing the <code class="docutils literal notranslate"><span class="pre">ModelWeightsLoader</span></code> object, the file format will be derived from <code class="docutils literal notranslate"><span class="pre">model_dir</span></code> through <code class="docutils literal notranslate"><span class="pre">detect_format</span></code>. The following formats are supported for now:</p>
|
||
<ul class="simple">
|
||
<li><p>Directory contains or file named <code class="docutils literal notranslate"><span class="pre">*.safetensors</span></code> (Recommended, has better performance)</p></li>
|
||
<li><p>Directory contains or file named <code class="docutils literal notranslate"><span class="pre">*.bin</span></code></p></li>
|
||
<li><p>Directory contains or file named <code class="docutils literal notranslate"><span class="pre">*.pth</span></code></p></li>
|
||
</ul>
|
||
<p>To support other formats or in-memory loaded models, the format need to be claimed in <code class="docutils literal notranslate"><span class="pre">ModelWeightsFormat</span></code>, <code class="docutils literal notranslate"><span class="pre">detect_format()</span></code>, <code class="docutils literal notranslate"><span class="pre">preload()</span></code> and <code class="docutils literal notranslate"><span class="pre">load_tensor()</span></code>.</p>
|
||
</section>
|
||
<section id="postprocessing-functions">
|
||
<h3>Postprocessing functions<a class="headerlink" href="#postprocessing-functions" title="Link to this heading"></a></h3>
|
||
<p>After translation and loading, a TRT-LLM key will become a tensor or a list of tensors, which is the input of postprocessing functions. <br />
|
||
Operations including QKV concatenating, MoE weight stacking and weight-only quantization can be handled here.
|
||
The template of postprocessing function is:</p>
|
||
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Example for 1-1 weights mapping</span>
|
||
<span class="k">class</span> <span class="nc">CustomizedModuleA</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span>
|
||
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="o">...</span><span class="p">):</span>
|
||
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="o">...</span><span class="p">)</span>
|
||
<span class="o">...</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">tp_dim</span> <span class="o">=</span> <span class="mi">0</span> <span class="c1"># Need to set or inherit from parent class</span>
|
||
|
||
<span class="k">def</span> <span class="nf">postprocess</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">tllm_key</span><span class="p">,</span> <span class="n">weights</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
|
||
<span class="n">weights</span> <span class="o">=</span> <span class="n">proc</span><span class="p">(</span><span class="n">weights</span><span class="p">)</span>
|
||
<span class="k">return</span> <span class="p">{</span><span class="n">tllm_key</span><span class="p">:</span> <span class="n">weights</span><span class="p">}</span>
|
||
|
||
<span class="c1"># Example for multiple-multiple weights mapping</span>
|
||
<span class="k">class</span> <span class="nc">CustomizedModuleB</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span>
|
||
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="o">...</span><span class="p">):</span>
|
||
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="o">...</span><span class="p">)</span>
|
||
<span class="o">...</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">tp_dim</span> <span class="o">=</span> <span class="mi">0</span> <span class="c1"># Need to set or inherit from parent class</span>
|
||
<span class="c1"># The default value of "weights" in tllm_to_externel_key_dict will be override</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">tllm_to_externel_key_dict</span> <span class="o">=</span> <span class="p">{</span><span class="s2">"weight"</span><span class="p">:</span> <span class="p">[</span><span class="s2">"qweight"</span><span class="p">,</span> <span class="s2">"qzeros"</span><span class="p">,</span> <span class="s2">"scales"</span><span class="p">]}</span>
|
||
|
||
<span class="k">def</span> <span class="nf">postprocess</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">tllm_key</span><span class="p">,</span> <span class="n">weights</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
|
||
<span class="c1"># Skipped the postprocess of zeros and weights_scaling_factor</span>
|
||
<span class="c1"># They are loaded in the postprocess of weight</span>
|
||
<span class="n">config</span> <span class="o">=</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"config"</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span> <span class="c1"># Passed through kwargs by default</span>
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="n">tllm_key</span><span class="o">.</span><span class="n">endswith</span><span class="p">(</span><span class="s2">"weight"</span><span class="p">):</span>
|
||
<span class="k">return</span> <span class="p">{}</span>
|
||
<span class="c1"># The order in weights is defined in tllm_to_externel_key_dict</span>
|
||
<span class="n">qweight</span><span class="p">,</span> <span class="n">qzeros</span><span class="p">,</span> <span class="n">scales</span> <span class="o">=</span> <span class="n">weights</span>
|
||
<span class="n">proccessed_weight</span><span class="p">,</span> <span class="n">proccessed_zeros</span> <span class="o">=</span> <span class="n">proc</span><span class="p">(</span><span class="n">qweight</span><span class="p">,</span> <span class="n">qzeros</span><span class="p">,</span> <span class="n">config</span><span class="o">.</span><span class="n">num_heads</span><span class="p">)</span>
|
||
<span class="k">return</span> <span class="p">{</span>
|
||
<span class="n">tllm_key</span><span class="p">:</span> <span class="n">proccessed_weight</span><span class="p">,</span>
|
||
<span class="n">tllm_key</span><span class="o">.</span><span class="n">replace</span><span class="p">(</span><span class="s2">"weight"</span><span class="p">,</span> <span class="s2">"zeros"</span><span class="p">):</span> <span class="n">proccessed_zeros</span><span class="p">,</span>
|
||
<span class="n">tllm_key</span><span class="o">.</span><span class="n">replace</span><span class="p">(</span><span class="s2">"weight"</span><span class="p">,</span> <span class="s2">"weights_scaling_factor"</span><span class="p">):</span> <span class="n">scales</span><span class="p">,</span>
|
||
<span class="p">}</span>
|
||
</pre></div>
|
||
</div>
|
||
</section>
|
||
</section>
|
||
<section id="examples">
|
||
<h2>Examples<a class="headerlink" href="#examples" title="Link to this heading"></a></h2>
|
||
<p>The <code class="docutils literal notranslate"><span class="pre">ModelWeightsLoader</span></code> class can support different models with the following levels:</p>
|
||
<section id="natively-supported-models">
|
||
<h3>Natively supported models<a class="headerlink" href="#natively-supported-models" title="Link to this heading"></a></h3>
|
||
<p>For models with native support, users can call the default weight loader without any other operations.</p>
|
||
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Using the model weights loader for LLaMA</span>
|
||
<span class="kn">from</span> <span class="nn">tensorrt_llm.models.model_weights_loader</span> <span class="kn">import</span> <span class="n">ModelWeightsLoader</span>
|
||
<span class="n">loader</span> <span class="o">=</span> <span class="n">ModelWeightsLoader</span><span class="p">(</span><span class="n">external_checkpoint_dir</span><span class="p">)</span>
|
||
<span class="n">loader</span><span class="o">.</span><span class="n">generate_tllm_weights</span><span class="p">(</span><span class="n">trtllm_model</span><span class="p">)</span>
|
||
</pre></div>
|
||
</div>
|
||
<p>For calibration-free quantization precisions, passing a properly quantized <code class="docutils literal notranslate"><span class="pre">trtllm_model</span></code> will let the weight loader load at the given precision accordingly. The configurations will be read from <code class="docutils literal notranslate"><span class="pre">trtllm_model.config</span></code> automatically. For now, LLaMA family models using the default <code class="docutils literal notranslate"><span class="pre">tllm_to_externel_key_dict</span></code> is supported natively.</p>
|
||
</section>
|
||
<section id="models-with-customized-key-names">
|
||
<h3>Models with customized key names<a class="headerlink" href="#models-with-customized-key-names" title="Link to this heading"></a></h3>
|
||
<p>For models with different naming logic, users can still call the default weight loader with <code class="docutils literal notranslate"><span class="pre">customized_key_dict</span></code> specified.</p>
|
||
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Using the model weights loader for the LLM part of LLaVA</span>
|
||
<span class="kn">from</span> <span class="nn">tensorrt_llm.models.model_weights_loader</span> <span class="kn">import</span> <span class="n">ModelWeightsLoader</span>
|
||
<span class="n">llava_dict</span> <span class="o">=</span> <span class="p">{</span>
|
||
<span class="s2">"transformer"</span><span class="p">:</span> <span class="s2">"language_model.model"</span><span class="p">,</span>
|
||
<span class="s2">"lm_head"</span><span class="p">:</span> <span class="s2">"language_model.lm_head"</span>
|
||
<span class="p">}</span>
|
||
<span class="n">loader</span> <span class="o">=</span> <span class="n">ModelWeightsLoader</span><span class="p">(</span><span class="n">external_checkpoint_dir</span><span class="p">,</span> <span class="n">llava_dict</span><span class="p">)</span>
|
||
<span class="n">loader</span><span class="o">.</span><span class="n">generate_tllm_weights</span><span class="p">(</span><span class="n">trtllm_model</span><span class="p">)</span>
|
||
</pre></div>
|
||
</div>
|
||
<p>Users need to specify the different part from the default <code class="docutils literal notranslate"><span class="pre">tllm_to_externel_key_dict</span></code>. The loader still have support across different precisions.
|
||
The support for LLaVA and Exaone is in <code class="docutils literal notranslate"><span class="pre">LLaMAForCausalLM.from_hugging_face()</span></code> of <a class="reference download internal" download="" href="../_downloads/408e9af6e2b04a79e78215bde246e8bc/model.py"><span class="xref download myst">model.py</span></a>, and can also be taken as examples.</p>
|
||
</section>
|
||
<section id="models-with-customized-weight-layout">
|
||
<h3>Models with customized weight layout<a class="headerlink" href="#models-with-customized-weight-layout" title="Link to this heading"></a></h3>
|
||
<p>For models with different weight layout, users can write the conversion loop explicitly and do customized operations.</p>
|
||
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Using the model weights loader for BLOOM</span>
|
||
<span class="kn">from</span> <span class="nn">tensorrt_llm.models.model_weights_loader</span> <span class="kn">import</span> <span class="n">ModelWeightsLoader</span>
|
||
<span class="n">bloom_dict</span> <span class="o">=</span> <span class="p">{</span>
|
||
<span class="s2">"transformer"</span><span class="p">:</span> <span class="s2">""</span><span class="p">,</span>
|
||
<span class="s2">"layers"</span><span class="p">:</span> <span class="s2">"h"</span><span class="p">,</span>
|
||
<span class="s2">"ln_f"</span><span class="p">:</span> <span class="s2">"ln_f"</span><span class="p">,</span>
|
||
<span class="s2">"lm_head"</span><span class="p">:</span> <span class="s2">"word_embeddings"</span><span class="p">,</span>
|
||
<span class="s2">"ln_embed"</span><span class="p">:</span> <span class="s2">"word_embeddings_layernorm"</span><span class="p">,</span>
|
||
<span class="s2">"vocab_embedding"</span><span class="p">:</span> <span class="s2">"word_embeddings"</span><span class="p">,</span>
|
||
<span class="s2">"attention"</span><span class="p">:</span> <span class="s2">"self_attention"</span><span class="p">,</span>
|
||
<span class="s2">"qkv"</span><span class="p">:</span> <span class="s2">"query_key_value"</span><span class="p">,</span>
|
||
<span class="s2">"dense"</span><span class="p">:</span> <span class="s2">"dense"</span><span class="p">,</span>
|
||
<span class="s2">"fc"</span><span class="p">:</span> <span class="s2">"dense_h_to_4h"</span><span class="p">,</span>
|
||
<span class="s2">"proj"</span><span class="p">:</span> <span class="s2">"dense_4h_to_h"</span><span class="p">,</span>
|
||
<span class="s2">"post_layernorm"</span><span class="p">:</span> <span class="s2">"post_attention_layernorm"</span><span class="p">,</span>
|
||
<span class="p">}</span>
|
||
<span class="n">loader</span> <span class="o">=</span> <span class="n">ModelWeightsLoader</span><span class="p">(</span><span class="n">external_checkpoint_dir</span><span class="p">,</span> <span class="n">bloom_dict</span><span class="p">)</span>
|
||
<span class="c1"># See ModelWeightsLoader.generate_tllm_weights()</span>
|
||
<span class="n">loader</span><span class="o">.</span><span class="n">update_key_mapping</span><span class="p">(</span><span class="n">trtllm_model</span><span class="p">)</span>
|
||
<span class="n">tllm_weights</span> <span class="o">=</span> <span class="p">{}</span>
|
||
<span class="k">for</span> <span class="n">tllm_key</span><span class="p">,</span> <span class="n">_</span> <span class="ow">in</span> <span class="n">tqdm</span><span class="p">(</span><span class="n">trtllm_model</span><span class="o">.</span><span class="n">named_parameters</span><span class="p">()):</span>
|
||
<span class="k">if</span> <span class="n">tllm_key</span><span class="o">.</span><span class="n">endswith</span><span class="p">(</span><span class="s2">"qkv"</span><span class="p">):</span>
|
||
<span class="c1"># Passing the callable handle</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">tllm_key</span><span class="p">,</span> <span class="n">preprocess</span><span class="o">=</span><span class="n">customized_preprocess</span><span class="p">))</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">tllm_key</span><span class="p">))</span>
|
||
<span class="n">loader</span><span class="o">.</span><span class="n">fill</span><span class="p">(</span><span class="n">tllm_weights</span><span class="p">)</span>
|
||
</pre></div>
|
||
</div>
|
||
<p>This will apply <code class="docutils literal notranslate"><span class="pre">preprocess</span></code> after <code class="docutils literal notranslate"><span class="pre">load_tensor()</span></code> and before <code class="docutils literal notranslate"><span class="pre">postprocess</span></code>, and demonstrates how to convert the loaded shard into default HF layout. The loader still have support for precisions quantized from FP16/BF16 (e.g. INT8-wo/INT4-wo), the other precisions may require special operations, and can be addressed inside the <code class="docutils literal notranslate"><span class="pre">preprocess</span></code> function.
|
||
The support for Qwen-1 is in <code class="docutils literal notranslate"><span class="pre">QWenForCausalLM.from_hugging_face()</span></code> of <a class="reference download internal" download="" href="../_downloads/b6815cf245cc7dc7a26a6f727fdc2dc4/model.py"><span class="xref download myst">model.py</span></a>, and can also be taken as example.</p>
|
||
</section>
|
||
<section id="fully-customized">
|
||
<h3>Fully customized<a class="headerlink" href="#fully-customized" title="Link to this heading"></a></h3>
|
||
<p>If the model weights loader cannot satisfy the requirements, users can write the conversion loop totally on their own.</p>
|
||
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">tllm_weights</span> <span class="o">=</span> <span class="p">{}</span>
|
||
<span class="k">for</span> <span class="n">tllm_key</span><span class="p">,</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">tqdm</span><span class="p">(</span><span class="n">trtllm_model</span><span class="o">.</span><span class="n">named_parameters</span><span class="p">()):</span>
|
||
<span class="c1"># Load from external checkpoints</span>
|
||
<span class="c1"># The load_tensor() function can also be called here</span>
|
||
<span class="n">tensor</span> <span class="o">=</span> <span class="o">...</span>
|
||
<span class="c1"># Convert tensor and set the values according to the config</span>
|
||
<span class="k">if</span> <span class="n">trtllm_model</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">quant_algo</span> <span class="o">==</span> <span class="n">xxx</span><span class="p">:</span>
|
||
<span class="o">...</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="o">...</span>
|
||
<span class="n">param</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">tensor</span>
|
||
</pre></div>
|
||
</div>
|
||
<p>In this mode, every precision require user’s own support.</p>
|
||
</section>
|
||
</section>
|
||
<section id="trouble-shooting">
|
||
<h2>Trouble shooting<a class="headerlink" href="#trouble-shooting" title="Link to this heading"></a></h2>
|
||
<p>The weights loader is an experimental feature for now, and is enabled for LLaMA family models and Qwen models by default.</p>
|
||
<p>If users are encountered with failure caused by <code class="docutils literal notranslate"><span class="pre">ModelWeightsLoader</span></code>, a workaround is passing environmental variable <code class="docutils literal notranslate"><span class="pre">TRTLLM_DISABLE_UNIFIED_CONVERTER=1</span></code> to disable the model weights loader and fallback to the legacy path.</p>
|
||
<p>This workaround will be removed in future version after the LLaMA/Qwen weights conversion is stable.</p>
|
||
</section>
|
||
</section>
|
||
|
||
|
||
</div>
|
||
</div>
|
||
<footer>
|
||
|
||
<hr/>
|
||
|
||
<div role="contentinfo">
|
||
<jinja2.runtime.BlockReference object at 0x7da747c04710>
|
||
|
||
<div class="footer">
|
||
<p>
|
||
Copyright © 2024 NVIDIA Corporation
|
||
</p>
|
||
<p>
|
||
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/privacy-policy/" target="_blank" rel="noopener"
|
||
data-cms-ai="0">Privacy Policy</a> |
|
||
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/privacy-center/" target="_blank" rel="noopener"
|
||
data-cms-ai="0">Manage My Privacy</a> |
|
||
<a class="Link" href="https://www.nvidia.com/en-us/preferences/start/" target="_blank" rel="noopener"
|
||
data-cms-ai="0">Do Not Sell or Share My Data</a> |
|
||
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/terms-of-service/" target="_blank"
|
||
rel="noopener" data-cms-ai="0">Terms of Service</a> |
|
||
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/accessibility/" target="_blank" rel="noopener"
|
||
data-cms-ai="0">Accessibility</a> |
|
||
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/company-policies/" target="_blank"
|
||
rel="noopener" data-cms-ai="0">Corporate Policies</a> |
|
||
<a class="Link" href="https://www.nvidia.com/en-us/product-security/" target="_blank" rel="noopener"
|
||
data-cms-ai="0">Product Security</a> |
|
||
<a class="Link" href="https://www.nvidia.com/en-us/contact/" target="_blank" rel="noopener"
|
||
data-cms-ai="0">Contact</a>
|
||
</p>
|
||
</div>
|
||
|
||
|
||
</div>
|
||
|
||
|
||
|
||
</footer>
|
||
</div>
|
||
</div>
|
||
</section>
|
||
</div>
|
||
<script>
|
||
jQuery(function () {
|
||
SphinxRtdTheme.Navigation.enable(true);
|
||
});
|
||
</script>
|
||
|
||
</body>
|
||
</html> |